Deceptive prompts lure AI agents into decoys and dead ends, cutting attack success by up to 91% in tests and raising the cost of a successful attack 23-fold.
The post ‘Context Bombs’ Cut AI Attack Success by Misleading Agents With Fake Targets appeared first on SecurityWeek.
https://www.securityweek.com/context-bombs-cut-ai-attack-success-by-misleading-agents-with-fake-targets/
In two internal incidents, agents read routine coding instructions as permission to weaken monitoring, and one gave a false answer about it.
The post OpenAI Agents Silenced Security Team Alerts, Removed Misalignment Monitor appeared first on SecurityWeek.
https://www.securityweek.com/openai-agents-silenced-security-team-alerts-removed-misalignment-monitor/
The DHS said agencies should remove DeepSeek products as well as third-party products and services developed using DeepSeek from their IT systems.
The post China Defends DeepSeek After US Ban of Its AI From Government Systems appeared first on SecurityWeek.
https://www.securityweek.com/china-defends-deepseek-after-us-ban-of-its-ai-from-government-systems/
Tasked with writing an application for a US defense contractor, the model made the code 19% more likely to be hackable.
The post Chinese AI Model Writes More Hackable Code When Told It’s for US Contractor appeared first on SecurityWeek.
https://www.securityweek.com/chinese-ai-model-writes-more-hackable-code-when-told-its-for-us-contractor/
Fitzgerald’s is a rare case where an official employed by the state’s top law enforcement and legal office has been accused of making and distributing AI CSAM, although some cops have been arrested for the crime. Earlier this year, a Pennsylvania cop was arrested for making AI porn from photos on driver’s licenses, and more recently in Korea, a police officer was arrested for making AI porn, including CSAM, The Korea Times reported.
In New Jersey, the deputy attorney general has been charged with possession of CSAM, dissemination of photos or film of CSAM, and “criminal use of a communication facility,” detectives said.
According to his LinkedIn, Fitzgerald had served as deputy attorney general since 2016, preparing insurance fraud and insurance licensee cases for trial as a state prosecutor. New Jersey’s district attorney, Joe Khan, said in a statement that New Jersey is aligned with the DOJ in viewing AI CSAM as being as harmful as CSAM.
“No one is above the law, least of all someone entrusted with upholding it,” Khan said. “Our office will pursue these cases aggressively and hold offenders fully accountable, regardless of their profession or title.”
https://arstechnica.com/tech-policy/2026/10/thegentlemanpervert-producing-ai-csam-is-a-nj-prosecutor-say-cops/
Noteworthy stories that might have slipped under the radar: BofA adds 1,000 patching contractors, KGB RAT pairs data theft with jump scares, and Langflow flaw under attack.
The post In Other News: ICS-Sniper Attack, Hacking XWorm Malware Operators, $250k Chrome Exploit appeared first on SecurityWeek.
https://www.securityweek.com/in-other-news-ics-sniper-attack-hacking-xworm-malware-operators-250k-chrome-exploit/
Left without funds for a coding task, an OpenAI model spent test crypto from a stranger’s wallet and left it out of its report.
The post OpenAI Models Hacked Data Portals, Raided a Crypto Wallet to Finish Tasks appeared first on SecurityWeek.
https://www.securityweek.com/openai-models-hacked-data-portals-raided-a-crypto-wallet-to-finish-tasks/
In Booz Allen’s isolated OT lab, AI models met every objective across eight scenarios and reached an industrial control network from a perimeter compromise in just over 16 minutes.
The post AI Agents Navigate Unfamiliar OT Systems and Move Equipment in Minutes, Booz Allen Finds appeared first on SecurityWeek.
https://www.securityweek.com/ai-agents-navigate-unfamiliar-ot-systems-and-move-equipment-in-minutes-booz-allen-finds/
The cybersecurity startup plans to expand engineering and go-to-market teams in the US and UK.
The post Harmonic Security Raises $50 Million for Workforce AI Protection appeared first on SecurityWeek.
https://www.securityweek.com/harmonic-security-raises-50-million-for-workforce-ai-protection/
Google revealed Gemini 4 Argon a few weeks ago, saying that it needed to move cautiously toward a full release because of the potential for nefarious use. Google seems to have worked all that out pretty quickly, though. Argon is beginning its wide rollout today, but it’s only coming to paying users at this time, and even those on lower-tier plans won’t have full access.
Nothing about Gemini 4 Argon’s capabilities has changed. Google still points to the benchmarks it published late last month as evidence of the model’s advanced reasoning and agentic coding capabilities. The model is supposedly trained to refuse malicious instructions, and an automated misalignment monitor can stop processing if it detects an unintended workflow. Google claims Argon needs that because it’s so capable.
The company has a few new examples of how Argon has helped inside Google, where it has been used by thousands of Googlers. Argon has reportedly been instrumental in making YouTube’s comment system more efficient. Argon identified redundant emoji rendering that, when eliminated, dropped server load by 45 percent. Engineers also turned Gemini 4 loose on the proprietary drone physics simulator used in Project Wing, and the AI found critical bugs that had been in the codebase since 2018. Google also notes that Argon has advanced its robotics program, showing a measurable step increase in embodied reasoning.
https://arstechnica.com/ai/2026/10/gemini-4-argon-begins-rolling-out-in-gemini-app-and-antigravity-but-only-if-you-pay/